Overhead line power outage risk prediction method and device
By adopting the Transformer end-to-end neural network structure and attention mechanism in overhead line power outage prediction, combined with the batch normalization layer, an efficient power outage prediction model is built, which solves the problem of insufficient accuracy in the existing technology and achieves high-precision Power outage prediction effect.
Patent Information
- Application Number
- CN202210577238.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-25
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2042-05-25
AI Technical Summary
The prior art is difficult to achieve high-precision overhead power outage fault prediction. The shallow neural network structure is limited by the model parameter quantity and expression ability, making it difficult to fit and mine potential data features.
Using an end-to-end neural network structure based on Transformer, combined with Attention_Module and MLP_3 modules, an attention mechanism and batch normalization layer are introduced to build an efficient power outage prediction model, process structured information feature data of overhead lines, and improve prediction accuracy.
The accuracy and performance of overhead line power outage prediction is significantly improved, the convergence problem of deep network training is solved, and the high-precision power outage prediction is achieved through the cross entropy loss function and the category imbalance loss function optimization model.
Smart Images

Figure CN115034447B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power distribution network big data, and in particular to a method and device for predicting the risk of power outage of an overhead line. Background Art
[0002] With the development of the economy and society, the coverage of substations and distribution networks in the power grid system has been continuously expanding. The intrusion of disastrous weather such as strong winds, lightning, and haze on the safe operation of the power grid has increased accordingly. Under the influence of extreme weather, 10kV overhead distribution lines are more likely to trip, causing regional power grid line failures and shutdowns, affecting the normal production and life of residents.
[0003] With the increasing number of severe weather such as haze and ice, the normal operation of power grid equipment has been greatly affected. There is an urgent need to conduct in-depth analysis of meteorological disasters in power grids and seek preventive measures. There have also been many studies on the risk assessment and prevention of meteorological disasters in power grids. For example, the Chinese invention patent application with application publication number CN108596449A discloses a distribution network reliability prediction method that considers the impact of weather on the probability of distribution network failure, including the following steps: (1) analyzing the correlation between the probability of failure of distribution network components and weather variables; (2) selecting the main weather variables by taking into account the applicability and accuracy of the model; (3) forming a combined prediction method for the failure rate of distribution network components based on support vector machines and additional momentum neural networks; (4) combining the prediction results of the probability of failure of distribution network components and the calculation method of distribution network reliability to predict the reliability of the distribution network. It can predict the reliability of the distribution network based on the impact of different weather variables on the probability of failure of distribution network components.
[0004] However, the neural network structure it uses is a shallow small-scale neural network. Due to the limitations of the number of model parameters and expression capabilities, it is difficult for the network to fit and mine the potential characteristics of the data. Summary of the invention
[0005] The technical problem to be solved by the present invention is how to achieve high-precision prediction of overhead line power outage faults.
[0006] The present invention solves the above technical problems through the following technical means:
[0007] In a first aspect, an embodiment of the present invention adopts a method for predicting the risk of power outage of an overhead line, the method comprising:
[0008] Obtaining overhead line structured information feature data;
[0009] Using the trained power outage prediction model to process the overhead line structured information feature data to obtain a power outage prediction result;
[0010] Among them, the power outage prediction model includes a Backbone network, a Transformer_Encoder module, an Attention_Module network, a Transformer_Decoder module and an MLP_3 module. The overhead line structured information feature data is used as the input of the Backbone network, the output of the Backbone network is connected to the Transformer_Encoder module, the output of the Transformer_Encoder module is connected to the Attention_Module network, the output features of the Attention_Module network and the global representation features of the database are used as the input of the Transformer_Decoder module, the output of the Transformer_Decoder module is connected to the MLP_3 module, and the MLP_3 module outputs the power outage prediction result.
[0011] To solve the problem of predicting overhead power outages under severe convective weather, the present invention designs an end-to-end neural network structure based on Transformer as a power outage prediction model. Thanks to the ability of the Transformer structure to capture long-distance feature dependencies, the prediction accuracy of the network can be greatly improved. In addition, an additional attention mechanism is introduced into the network model to obtain features that are more relevant to the classification task, further improving the performance of the overhead power outage prediction model.
[0012] Furthermore, the overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data;
[0013] The meteorological characteristic data include the maximum value f1 of the severe convection radar intensity, the lightning current f2 and the number of lightning return strokes f3; the geographical characteristic data include the elevation l1, the slope l2 and the slope direction l3; the historical fault characteristic data of the power grid include the number of overloads n1, the number of heavy loads n2 and the number of defects n3.
[0014] Furthermore, after acquiring the overhead line structured information feature data, the method further includes:
[0015] Performing dimension-unified processing on the overhead line structured information characteristic data to obtain dimension-unified data;
[0016] Accordingly, the power outage prediction model is used to process the overhead line structured information feature data to obtain a power outage prediction result, specifically:
[0017] The power outage prediction model is used to process the dimensionally unified data to obtain a power outage prediction result.
[0018] Furthermore, the Backbone network includes a fully connected layer and a downsampling layer, the overhead line structured information feature data is used as the input of the fully connected layer, the output of the fully connected layer is connected to the downsampling layer, and the downsampling layer outputs the extracted feature F feat .
[0019] Furthermore, the Transformer_Encoder module includes a batch normalization layer Norm_1, a first multi-head attention layer, a batch normalization layer Norm_2, and a multi-layer neural network MLP_1. The extracted features output by the Backbone network are used as input of the batch normalization layer Norm_1. The output of the batch normalization layer Norm_1 is connected to the first multi-head attention layer. The input of the batch normalization layer Norm_2 is a first fused feature obtained by fusion of the output features of the first multi-head attention layer and the extracted features output by the Backbone network. The output of the batch normalization layer Norm_2 is connected to the multi-layer neural network MLP_1. The features output by the multi-layer neural network MLP_1 are fused with the first fused features to obtain enhanced features.
[0020] Furthermore, the Attention_Module network is used to fuse the enhanced features output by the Transformer_Encoder module to obtain features for classification tasks, specifically:
[0021] Updating the enhanced features using learnable weight parameters to obtain updated features;
[0022] The updated features are fused with the enhanced features to obtain the features used for the classification task.
[0023] Furthermore, the Transformer_Decoder module includes a batch normalization layer Norm_3, a second multi-head attention layer, a batch normalization layer Norm_4, and a multi-layer neural network MLP_2. The global representation feature of the database serves as the input of the batch normalization layer Norm_3. The output of the batch normalization layer Norm_3 and the output of the Attention_Module network are both connected to the second multi-head attention layer. The input of the batch normalization layer Norm_4 is the second fused feature obtained by fusing the output feature of the second multi-head attention layer with the global representation feature of the database. The output of the batch normalization layer Norm_4 is connected to the multi-layer neural network MLP_2. The output feature of the multi-layer neural network MLP_2 and the second fused feature are fused to obtain the updated global representation feature of the database. The updated global representation feature of the database serves as the input of the MLP_3 module.
[0024] Furthermore, the method further comprises:
[0025] Collect sample data and construct a sample set, wherein the sample data includes historical meteorological characteristic data, historical geographical characteristic data, historical power grid characteristic data and real fault labels;
[0026] The power outage prediction model is trained by using the sample set and adopting a back propagation algorithm to obtain the trained power outage prediction model.
[0027] Furthermore, the loss function used in the training process of the power outage prediction model is:
[0028] L final (P pre )=-α cls (1-P pre ) γ log(P pre )
[0029] Among them, P pre is the predicted probability; α cls is a piecewise function, N neg is a negative sample, N pos Positive sample, γ is the operation parameter, γ>1.
[0030] In addition, the present invention also proposes an overhead line power outage risk prediction device, the device comprising:
[0031] An acquisition module, used to acquire overhead line structured information characteristic data, wherein the overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data;
[0032] A prediction module, used to process the overhead line structured information feature data using a trained power outage prediction model to obtain a power outage prediction result;
[0033] Among them, the power outage prediction model includes a Backbone network, a Transformer_Encoder module, an Attention_Module network, a Transformer_Decoder module and an MLP_3 module. The overhead line structured information feature data is used as the input of the Backbone network, the output of the Backbone network is connected to the Transformer_Encoder module, the output of the Transformer_Encoder module is connected to the Attention_Module network, the output features of the Attention_Module network and the global representation features of the database are used as the input of the Transformer_Decoder module, the output of the Transformer_Decoder module is connected to the MLP_3 module, and the MLP_3 module outputs the power outage prediction result.
[0034] The advantages of the present invention are:
[0035] (1) Aiming at the problem of predicting overhead power outages under severe convective weather, the present invention designs an end-to-end neural network structure based on Transformer as a power outage prediction model. Thanks to the ability of the Transformer structure to capture long-distance feature dependencies, the prediction accuracy of the network can be greatly improved. In addition, an additional attention mechanism is introduced into the network model to obtain features that are more relevant to the classification task, further improving the performance of the overhead power outage prediction model.
[0036] (2) Batch normalization operations are used in the network structure to solve the deep network optimization problem and the training convergence problem of the power outage prediction model.
[0037] (3) By modeling the power outage prediction problem as a classification problem, the cross entropy loss function optimization model was introduced to achieve a high-precision overhead line power outage prediction model. And based on the characteristics of the dataset for this task, a loss function for category imbalance and difficult sample mining was proposed on the basis of the cross entropy loss function, which improved the effectiveness of the training process.
[0038] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 is a flow chart of a method for predicting the risk of an overhead line power outage in one embodiment of the present invention;
[0040] Figure 2 is a network structure diagram of a power outage prediction model in one embodiment of the present invention;
[0041] Figure 3 is a structural diagram of a Transformer_Encoder module in one embodiment of the present invention;
[0042] Figure 4 is a structural diagram of the Attention_Module network in one embodiment of the present invention;
[0043] Figure 5 is a structural diagram of a Transformer_Decoder module in one embodiment of the present invention;
[0044] Figure 6 It is a structural diagram of an overhead line power outage risk prediction device in one embodiment of the present invention. DETAILED DESCRIPTION
[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0046] like Figure 1 to Figure 2 As shown, the first embodiment of the present invention discloses a method for predicting the risk of power outage of overhead lines, the method comprising the following steps:
[0047] S10, obtaining overhead line structured information feature data;
[0048] S30, using the trained power outage prediction model to process the overhead line structured information feature data to obtain a power outage prediction result;
[0049] Among them, the power outage prediction model includes a Backbone network, a Transformer_Encoder module, an Attention_Module network, a Transformer_Decoder module and an MLP_3 module. The overhead line structured information feature data is used as the input of the Backbone network, the output of the Backbone network is connected to the Transformer_Encoder module, the output of the Transformer_Encoder module is connected to the Attention_Module network, the output features of the Attention_Module network and the global representation features of the database are used as the input of the Transformer_Decoder module, the output of the Transformer_Decoder module is connected to the MLP_3 module, and the MLP_3 module outputs the power outage prediction result.
[0050] To solve the problem of predicting overhead power outages under severe convective weather, this embodiment designs an end-to-end neural network structure based on Transformer as a power outage prediction model. Thanks to the ability of the Transformer structure to capture long-distance feature dependencies, the prediction accuracy of the network can be greatly improved. In addition, an additional attention mechanism is introduced into the network model to obtain features that are more relevant to the classification task, further improving the performance of the overhead power outage prediction model.
[0051] It should be noted that this embodiment combines historical status big data such as severe convective weather, geographical features and historical power grid faults to form structured information feature data of overhead lines, and uses the above historical big data to train and optimize large-scale deep neural networks to obtain the pre-trained power outage prediction model in this embodiment.
[0052] In one embodiment, the overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data;
[0053] The meteorological characteristic data include the maximum value f1 of the severe convection radar intensity, the lightning current f2 and the number of lightning return strokes f3; the geographical characteristic data include the elevation l1, the slope l2 and the slope direction l3; the historical fault characteristic data of the power grid include the number of overloads n1, the number of heavy loads n2 and the number of defects n3.
[0054] In one embodiment, since the dimensions of the overhead line structured information feature data are inconsistent and unfriendly to neural network training, the dimensions of the overhead line structured information feature data are unified, and the specific formula is as follows:
[0055] x′=(xx mean ) / x std
[0056] Among them, x′ is the variable value after dimension unification; x is the original variable value; x mean and x std are the mean and standard deviation of the original variables, respectively.
[0057] After the above-mentioned dimensional unification, the input data data = (f1, f2, f3, l1, l2, l3, n1, n2, n3)′ of the power outage prediction model is obtained.
[0058] In one embodiment, the Backbone network includes a fully connected layer and a downsampling layer, the overhead line structured information feature data is used as the input of the fully connected layer, the output of the fully connected layer is connected to the downsampling layer, and the downsampling layer outputs extracted features.
[0059] This embodiment uses the Backbone network to extract features from the input data to obtain deep features rich in semantic information, where Backbone uses a conventional fully connected layer and downsampling strategy to extract features. The process is expressed as:
[0060] F feat =f(W backbone ,data)
[0061] Where: W backbone is the parameter of the Backbone network, and f represents the feature extraction process.
[0062] In one embodiment, if Figure 3 As shown, the Transformer_Encoder module includes a batch normalization layer Norm_1, a first multi-head attention layer Muti-Head Attention_1, a batch normalization layer Norm_2, and a multi-layer neural network MLP_1. The extracted features output by the Backbone network are used as the input of the batch normalization layer Norm_1. The output of the batch normalization layer Norm_1 is connected to the first multi-head attention layer. The input of the batch normalization layer Norm_2 is the first fused feature obtained by fusion of the output features of the first multi-head attention layer and the extracted features output by the Backbone network. The output of the batch normalization layer Norm_2 is connected to the multi-layer neural network MLP_1. The features output by the multi-layer neural network MLP_1 are fused with the first fused features to obtain enhanced features.
[0063] It should be noted that in this embodiment, the Backbone network is used to extract the feature F featFinally, the Transformer_Encoder module is used to find the dependencies between the input features, and then the feature F is weighted based on its similarity. feat Enhance and obtain the enhanced feature F feat_after_encoder .
[0064] In one embodiment, if Figure 4 As shown, the Attention_Module network is used to fuse the enhanced features output by the Transformer_Encoder module to obtain features for classification tasks, specifically:
[0065] Updating the enhanced features using learnable weight parameters to obtain updated features;
[0066] The updated feature and the enhanced feature are fused to obtain the feature F used for the classification task. cls .
[0067] Specifically, the learnable weight parameter W = {w i |i=1,2…,n},F={f i |i=1,2,…,n}, use W to update the feature F, and get F′={f′ i =w i ·f i |i=1,2,…,n}, and finally perform feature fusion to obtain the final feature F used for classification cls =F+F′, F=F feat_after_encoder .
[0068] It should be noted that after the Transformer_Encoder module outputs the enhanced features, the Attention_Module network is used to enhance the features F feat_after_encoder Fusion is performed, so the feature F cls More suitable for classification tasks, an additional attention mechanism is introduced into the network model to obtain features that are more relevant to the classification tasks, further improving the performance of the overhead line power outage prediction model.
[0069] In one embodiment, if Figure 5As shown, the Transformer_Decoder module includes a batch normalization layer Norm_3, a second multi-head attention layer Muti-Head Attention_2, a batch normalization layer Norm_4, and a multi-layer neural network MLP_2. The database global representation feature is used as the input of the batch normalization layer Norm_3. The output of the batch normalization layer Norm_3 and the output of the Attention_Module network are both connected to the second multi-head attention layer. The input of the batch normalization layer Norm_4 is the second fused feature obtained by fusing the output feature of the second multi-head attention layer with the global representation feature of the database. The output of the batch normalization layer Norm_4 is connected to the multi-layer neural network MLP_2. The output feature of the multi-layer neural network MLP_2 and the second fused feature are fused to obtain an updated database global representation feature class_token_1. The updated database global representation feature class_token_1 is used as the input of the MLP_3 module.
[0070] It should be noted that, in order to perform end-to-end prediction, this embodiment introduces a feature class_token, which is a global representation of a database feature. By using the Transformer_Decoder module to calculate class_token and F cls The similarity of F cls Learn the features related to classification and get the updated class_token_1.
[0071] Furthermore, the updated class_token_1 is feature mapped through the MLP_3 module to obtain the final classification score.
[0072] Specifically, the MLP_1, MLP_2 and MLP_3 modules are all composed of simple linear mapping layers and normalization layers, and are universal modules.
[0073] Furthermore, the power outage prediction model outputs a power outage confidence score, and the output power outage confidence score is compared with the normal confidence score to determine whether a power outage occurs, specifically:
[0074]
[0075] Furthermore, this embodiment designs a reasonable risk level mechanism according to the risk probability output by the model, predicts the risk level of the region, and finally obtains three types of power outage prediction results: low risk, medium risk, and high risk.
[0076] In one embodiment, the method further comprises:
[0077] Collect sample data and construct a sample set, wherein the sample data includes historical meteorological characteristic data, historical geographical characteristic data, historical power grid characteristic data and real fault labels;
[0078] The power outage prediction model is trained by using the sample set and adopting a back propagation algorithm to obtain the trained power outage prediction model.
[0079] It should be noted that since the training data for the power outage prediction task has tens of millions of items, as time accumulates, it will soon reach over 100 million items. Faced with such a large amount of data information, traditional shallow small-scale neural networks (such as the classic three-layer artificial neural network) are limited by the number of model parameters and expression capabilities. It is difficult for the network to fit and mine the potential features of the data. However, directly increasing the number of neural network layers or width will cause the gradient to disappear during the training process, resulting in the inability to optimize the neural network.
[0080] Specifically, according to common sense, the solution space of a deeper network structure includes the solution space of a shallow network structure, which means that a deeper network structure can obtain a better solution and perform better than a shallow network. However, this is not actually the case. Whether in terms of training error or test error, a deep network may have a worse error than a shallow network, which also proves that it is not due to overfitting. The reason for this phenomenon may be that the stochastic gradient descent strategy often does not solve the global optimal solution, but the local optimal solution. Since the structure of a deep network is more complex, the gradient descent algorithm is more likely to obtain a local optimal solution.
[0081] This embodiment can perform end-to-end training by specifically designing the Transformer model structure, without having to divide the model into several different stages for training. The neural network is trained using the collected historical state data and supervised using data labels to complete end-to-end training.
[0082] In one embodiment, the method further comprises:
[0083] After the sample data is dimensionally unified, it is used to train a neural network model.
[0084] In one embodiment, the method further includes introducing batch normalization during the training process to perform batch normalization on the features, which can better solve the gradient vanishing problem, specifically:
[0085]
[0086]
[0087]
[0088] Where: is the mean value of the data, x (i) is the original value of a specific data. is the variance of the data set, is the normalized value of a specific data, ∈ is the regularization coefficient to prevent the denominator from being 0, and m is the total number of samples in the data set.
[0089] In one embodiment, the loss function used in the training process of the power outage prediction model is:
[0090] L final (P pre )=-α cls (1-P pre ) γ log(P pre )
[0091] Among them, P pre is the predicted probability; α cls is a piecewise function, N neg is a negative sample, N pos Positive sample, γ is the operation parameter, γ>1.
[0092] Specifically, since the power outage prediction problem is modeled as a binary classification problem in this embodiment, namely, the power outage category and the normal category, the cross entropy loss function can be used to optimize the neural network parameters. In the case of binary classification, the model finally needs to predict only two cases. The probability of prediction for each category is p and 1-p, and the cross entropy loss function is:
[0093]
[0094] in, Representation sample The true fault label of the positive class is 1 and the negative class is 0. Representation sample The probability of predicting the positive class, N represents the total number of samples.
[0095] Specifically for this task, the last layer of the power outage prediction model obtains the score of each category, namely the power outage score and the normal score; the score is passed through the sigmoid (or softmax) function to obtain the probability output; the cross entropy loss function is calculated between the category probability output predicted by the model and the one hot form of the true category.
[0096] Furthermore, the construction of the category imbalance and difficult sample mining loss function. During the model training process, in order to solve the problem of category imbalance in the data set, a category imbalance loss function was introduced based on the original classification loss. First, the basic classification loss function was simplified:
[0097]
[0098]
[0099] make Then we have:
[0100] L bce (p opt ,y)=-log(P pre );
[0101] P opt is the category probability predicted by the designed neural network model, and y is the sample.
[0102] To solve the problem of imbalance between positive and negative samples, set the piecewise function α is 0-1.
[0103] Then we have:
[0104] L bce (p opt ,y)=-α cls log(P pre ).
[0105] According to this formula, the loss weight of samples with a small number can be amplified, while the loss weight of categories with a large number can be reduced, and dynamic adjustment can be performed. If the number of positive samples is small, the calculated probability will be given a higher weight, and if the number of negative samples is small, a lower weight will be given to balance the contribution of positive and negative samples, solving the problem of imbalance in the number of positive and negative samples.
[0106] In order to solve the problem of difficult sample mining, this embodiment further processes the loss function and constructs the following loss function:
[0107] L hard (P pre )=-(1-P pre ) γ log(P pre )
[0108] Through this loss function, we can find that for a positive sample, if its predicted probability P pre =0.8, indicating that the prediction is correct, so it is an easy sample, (1-P pre ) γ=(0.2) γ , let γ = 2, then the weight of this sample is 0.04, which does not contribute much to the final loss function weight. On the contrary, if P pre =0.1, the contribution to the final loss function is 0.81, which means it plays the role of difficult sample mining and needs to focus on difficult samples.
[0109] The loss function of the task of this patent is finally unified as L final (P pre )=-α cls (1-P pre ) γ log(P pre ).
[0110] This embodiment designs an end-to-end neural network model based on Transformer for the optimization problem of the overhead power outage prediction model. By modeling the power outage prediction problem as a classification problem, a cross entropy loss function optimization model is introduced to achieve a high-precision overhead power outage prediction model. In addition, based on the characteristics of the dataset of this task, a loss function for class imbalance and difficult sample mining is proposed on the basis of the cross entropy loss function, which improves the effectiveness of the training process.
[0111] In addition, if Figure 6 As shown, the second embodiment of the present invention provides an overhead line power outage risk prediction device, the device comprising:
[0112] An acquisition module 10 is used to acquire overhead line structured information characteristic data, wherein the overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data;
[0113] A prediction module 20, configured to process the overhead line structured information feature data using a trained power outage prediction model to obtain a power outage prediction result;
[0114] Among them, the power outage prediction model includes a Backbone network, a Transformer_Encoder module, an Attention_Module network, a Transformer_Decoder module and an MLP_3 module. The overhead line structured information feature data is used as the input of the Backbone network, the output of the Backbone network is connected to the Transformer_Encoder module, the output of the Transformer_Encoder module is connected to the Attention_Module network, the output features of the Attention_Module network and the global representation features of the database are used as the input of the Transformer_Decoder module, the output of the Transformer_Decoder module is connected to the MLP_3 module, and the MLP_3 module outputs the power outage prediction result.
[0115] To solve the problem of predicting overhead power outages under severe convective weather, this embodiment designs an end-to-end neural network structure based on Transformer as a power outage prediction model. Thanks to the ability of the Transformer structure to capture long-distance feature dependencies, the prediction accuracy of the network can be greatly improved. In addition, an additional attention mechanism is introduced into the network model to obtain features that are more relevant to the classification task, further improving the performance of the overhead power outage prediction model.
[0116] In one embodiment, the overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data;
[0117] The meteorological characteristic data include the maximum value f1 of the severe convection radar intensity, the lightning current f2 and the number of lightning return strokes f3; the geographical characteristic data include the elevation l1, the slope l2 and the slope direction l3; the historical fault characteristic data of the power grid include the number of overloads n1, the number of heavy loads n2 and the number of defects n3.
[0118] In one embodiment, the device further includes a dimension unification module for unifying the dimension of the overhead line structured information feature data. The specific formula is as follows:
[0119] x′=(xx mean ) / x std
[0120] Among them, x′ is the variable value after dimension unification; x is the original variable value; x mean and x std are the mean and standard deviation of the original variables, respectively.
[0121] After the above-mentioned dimensional unification, the input data data = (f1, f2, f3, l1, l2, l3, n1, n2, n3)′ of the power outage prediction model is obtained.
[0122] In one embodiment, the apparatus further comprises a training module, which is used to:
[0123] Collect sample data and construct a sample set, wherein the sample data includes historical meteorological characteristic data, historical geographical characteristic data, historical power grid characteristic data and real fault labels;
[0124] The power outage prediction model is trained by using the sample set and adopting a back propagation algorithm to obtain the trained power outage prediction model.
[0125] This embodiment can perform end-to-end training by specifically designing the Transformer model structure, without having to divide the model into several different stages for training. The neural network is trained using the collected historical state data and supervised using data labels to complete end-to-end training.
[0126] In one embodiment, the loss function used by the training module during the training process is:
[0127] L final (P pre )=-α cls (1-P pre ) γ log(P pre )
[0128] Among them, P pre is the predicted probability; α cls is a piecewise function, N neg is a negative sample, N pos Positive sample, γ is the operation parameter, γ>1.
[0129] It should be noted that other embodiments of the overhead line power outage risk prediction device of the present invention or implementation methods can refer to the above-mentioned method embodiments, which will not be repeated here.
[0130] It should be noted that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.
[0131] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0132] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0133] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0134] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.
Claims
1. A method for predicting the risk of an overhead line power outage, characterized in that: The method comprises: Obtaining overhead line structured information feature data; Using the trained power outage prediction model to process the overhead line structured information feature data to obtain a power outage prediction result; The power outage prediction model includes a Backbone network, a Transformer_Encoder module, an Attention_Module network, a Transformer_Decoder module and an MLP_3 module. The overhead line structured information feature data is used as the input of the Backbone network. The output of the Backbone network is connected to the Transformer_Encoder module. The output of the Transformer_Encoder module is connected to the Attention_Module network. The output features of the Attention_Module network and the global representation features of the database are used as the input of the Transformer_Decoder module. The output of the Transformer_Decoder module is connected to the MLP_3 module. The MLP_3 module outputs the power outage prediction result. The loss function used in the training process of the power outage prediction model is: in, is the predicted probability; is a piecewise function, , , is a negative sample, Positive sample, is the operation parameter, >1, For the sample, is 0-1.
2. The overhead line power outage risk prediction method according to claim 1, characterized in that: The overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data; The meteorological characteristic data include the maximum value f1 of the severe convection radar intensity, the lightning current f2 and the number of lightning return strokes f3; the geographical characteristic data include the elevation l1, the slope l2 and the slope direction l3; the historical fault characteristic data of the power grid include the number of overloads n1, the number of heavy loads n2 and the number of defects n3.
3. The overhead line power outage risk prediction method according to claim 1, characterized in that: After acquiring the overhead line structured information characteristic data, the method further includes: Performing dimension-unified processing on the overhead line structured information characteristic data to obtain dimension-unified data; Accordingly, the power outage prediction model is used to process the overhead line structured information feature data to obtain a power outage prediction result, specifically: The power outage prediction model is used to process the dimensionally unified data to obtain a power outage prediction result.
4. The overhead line power outage risk prediction method according to claim 1, characterized in that: The Backbone network includes a fully connected layer and a downsampling layer. The overhead line structured information feature data is used as the input of the fully connected layer. The output of the fully connected layer is connected to the downsampling layer, and the downsampling layer outputs extracted features.
5. The overhead line power outage risk prediction method according to claim 1, characterized in that: The Transformer_Encoder module includes a batch normalization layer Norm_1, a first multi-head attention layer, a batch normalization layer Norm_2, and a multi-layer neural network MLP_1. The extracted features output by the Backbone network are used as the input of the batch normalization layer Norm_1. The output of the batch normalization layer Norm_1 is connected to the first multi-head attention layer. The input of the batch normalization layer Norm_2 is the first fused feature obtained by fusion of the output features of the first multi-head attention layer and the extracted features output by the Backbone network. The output of the batch normalization layer Norm_2 is connected to the multi-layer neural network MLP_1. The features output by the multi-layer neural network MLP_1 are fused with the first fused features to obtain enhanced features.
6. The overhead line power outage risk prediction method according to claim 1, characterized in that: The Attention_Module network is used to fuse the enhanced features output by the Transformer_Encoder module to obtain features for classification tasks, specifically: Updating the enhanced features using learnable weight parameters to obtain updated features; The updated features are fused with the enhanced features to obtain the features used for the classification task.
7. The overhead line power outage risk prediction method according to claim 1, characterized in that: The Transformer_Decoder module includes a batch normalization layer Norm_3, a second multi-head attention layer, a batch normalization layer Norm_4, and a multi-layer neural network MLP_2. The global representation feature of the database serves as the input of the batch normalization layer Norm_3. The output of the batch normalization layer Norm_3 and the output of the Attention_Module network are both connected to the second multi-head attention layer. The input of the batch normalization layer Norm_4 is the second fused feature obtained by fusing the output feature of the second multi-head attention layer with the global representation feature of the database. The output of the batch normalization layer Norm_4 is connected to the multi-layer neural network MLP_2. The output feature of the multi-layer neural network MLP_2 and the second fused feature are fused to obtain the updated global representation feature of the database. The updated global representation feature of the database serves as the input of the MLP_3 module.
8. The overhead line power outage risk prediction method according to claim 1, characterized in that: The method further comprises: Collect sample data and construct a sample set, wherein the sample data includes historical meteorological characteristic data, historical geographical characteristic data, historical power grid characteristic data and real fault labels; The power outage prediction model is trained by using the sample set and adopting a back propagation algorithm to obtain the trained power outage prediction model.
9. An overhead line power outage risk prediction device, characterized in that: The device comprises: An acquisition module, used to acquire overhead line structured information characteristic data, wherein the overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data; A prediction module, used to process the overhead line structured information feature data using a trained power outage prediction model to obtain a power outage prediction result; The power outage prediction model includes a Backbone network, a Transformer_Encoder module, an Attention_Module network, a Transformer_Decoder module and an MLP_3 module. The overhead line structured information feature data is used as the input of the Backbone network. The output of the Backbone network is connected to the Transformer_Encoder module. The output of the Transformer_Encoder module is connected to the Attention_Module network. The output features of the Attention_Module network and the global representation features of the database are used as the input of the Transformer_Decoder module. The output of the Transformer_Decoder module is connected to the MLP_3 module. The MLP_3 module outputs the power outage prediction result. The loss function used in the training process of the power outage prediction model is: in, is the predicted probability; is a piecewise function, , , is a negative sample, Positive sample, is the operation parameter, >1, For the sample, is 0-1.
10. The overhead line power outage risk prediction device according to claim 9, characterized in that: The overhead line structured information characteristic data includes meteorological characteristic data, geographical characteristic data and power grid historical fault characteristic data; The meteorological characteristic data include the maximum value f1 of the severe convection radar intensity, the lightning current f2 and the number of lightning return strokes f3; the geographical characteristic data include the elevation l1, the slope l2 and the slope direction l3; the historical fault characteristic data of the power grid include the number of overloads n1, the number of heavy loads n2 and the number of defects n3.
Citation Information
Patent Citations
Distribution network reliability prediction method of considering influences of weather on distribution network fault probability
CN108596449A
Commodity sales volume prediction method and device based on Transformer + LSTM neural network model
CN111626764A
Bearing fault diagnosis method and system based on Transform and data enhancement
CN114036978A
Communication network traffic prediction method and system, storage medium and computer equipment
CN114422381A
Risk control method, apparatus, electronic device, and computer-readable storage medium
WO2021169115A1
